On creation of reference image for degraded documents binarisation
Soharab Hossain Shaikh, Ayan Dey, Khalid Saeed, Nabendu Chaki · International Journal of Computational Vision and Robotics · 2015
In degraded document image processing, a reference image is required for quantitative evaluation. Most of the papers found in the literature rely on creating reference images based on manual processing which is often subjective and prone to error. This paper proposes a method for creating reference image for degraded documents binarisation. Another standard approach towards automated creation of reference image is majority voting. The reference image produced by the majority voting scheme, does not discriminate the wrong estimation of threshold from the correct one by considering the possibility of strong bias due to poor estimation of threshold by some methods for a particular image. This leads to bad estimation of threshold for reference image. The proposed method overcomes the biasness of majority voting to a great extent by discarding some of the binarisation algorithms participating in the creation of the final reference image. This method gives a partially automated approach in calculating a threshold for the reference image depending on the type of degradation present in the input image. Experimental verification on standard benchmark images illustrates the effectiveness of the proposed method over majority voting approach.